via-camera upstream test results
corbet at lwn.net
Thu Oct 28 15:05:02 EDT 2010
On Wed, 27 Oct 2010 16:23:10 +0100
Daniel Drake <dsd at laptop.org> wrote:
> Testing with clean linux-next as of today. 2 via-camera issues:
If that's all that goes wrong we've come a fair ways...:)
> 1. Approx 13 "VIA DMA timeout!" messages are printed every second
That seems undesirable. In the past there was trouble with the
user-space xv code mucking around with the interrupt control registers,
but I *thought* that had been fixed. One could try lengthening the
msleep() call right before that message (it's in via-core.c), but that
shouldn't really be necessary.
> 2. If you start a capture with "gst-launch v4l2src ! xvimagesink" then
> change to a VT console then back to X, the display stops updating
> properly - it seems to have locked itself onto a few frames from
> before the VT switch happened, and plays them in an endless loop
> Also happens with "gst-launch v4l2src ! ffmpegcolorspace ! ximagesink"
I thought I had tried that, but, evidently not. Something in the
framebuffer switch code is clearly interfering with the video capture
engine, it would be nice to figure out what that is. If you're
capturing to a file, you'd really want that to survive a switch.
This will be something deep in the viafb code. Will take a bit of
digging; I don't know much about how VT switch actually works. My
guess is it's mucking with a register it shouldn't be.
I can take a look at both, but not before KS/Plumbers.
More information about the Devel